Meta has unveiled a cutting-edge artificial intelligence technology that decodes brain waves and translates them into text, potentially changing the lives of millions of people with paralysis and speech impairments.
On Monday, the company announced Brain2Qwerty v2, a system that can be described as a rudimentary form of "mind-reading" using algorithms . Although the research is still in its early stages, it opens the door to a future that may not be far off, where patients suffering from conditions such as aphasia, locked-in syndrome, amyotrophic lateral sclerosis (ALS), and other paralyzing neurological disorders could communicate simply by thinking, without the need for complex and expensive brain implant surgeries.
Meta wrote in its announcement: "We believe this research has the potential to make a real difference to millions of people who suffer from brain lesions that prevent them from communicating."
The company decided to make the basic programming code for the new system, as well as for the previous version, available for free on the internet, with the aim of accelerating the pace of scientific discoveries in the field of neuroscience, by enabling researchers around the world to access and develop this technology.
To train the new model, the researchers collaborated with the Basque Center for Cognition, Brain and Language in San Sebastián, Spain. Nine healthy volunteers, aged 25 to 56, participated in the trials, and were asked to write more than 2,500 sentences over ten sessions.
During these sessions, their brain activity wasmonitored using magnetoencephalography (MEG), which measures the minute electrical fields generated by neural activity. Written sentences and brain scans were then used as raw training data for the system.
The new system achieved remarkable accuracy in decoding brain activity, with a word-level accuracy rate of 78%, meaning that more than half of the sentences decoded from brainwaves contained only one grammatical error. This is a significant improvement over the previous version, v1, which achieved only 48% accuracy.
The researchers also discovered that the system's accuracy increases as the amount of training data provided to it increases, suggesting that applying simple measurement laws could lead to the construction of more capable systems in the future.
The researchers wrote: "If extensive training on non-invasive magnetic resonance electroencephalography (MEG - advanced non-surgical neuroimaging technology) data can eventually eliminate the need for neurosurgery, it will represent a radical transformation in patient care."
The new system relies on the same pattern recognition technology used in popular chatbots like ChatGPT and Meta's Llama. The decryption process is carried out in stages:
Phase 1: Brain waves measured by artificial intelligence are translated into symbols representing individual letters.
Phase Two: Another artificial intelligence system organizes these letters into complete words.
Stage three: A large linguistic model takes over, transforming the chaos of letters and words into coherent and understandable sentences.
This is the first time a large linguistic model has been successfully used to translate chaotic brain activity into organized sentences, providing a valuable model for future researchers in the field of brain-machine interfaces.
In addition to its multi-level decoding system, Brain2Qwerty also relies on a set of AI-powered "automated search agents" whose task is to automatically and independently improve the decoding process to enhance its accuracy and efficiency. These agents have been trained to "repeatedly modify the codebase to generate new and better constructs," resulting in a significant improvement in the word-level error rate.
